[deb_ffmpeg.git] / ffmpeg / libavcodec / ppc / hpeldsp_altivec.c

/*
 * Copyright (c) 2002 Brian Foley
 * Copyright (c) 2002 Dieter Shirley
 * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "config.h"

#if HAVE_ALTIVEC_H
#include <altivec.h>
#endif

#include "libavutil/attributes.h"
#include "libavutil/cpu.h"
#include "libavutil/ppc/cpu.h"
#include "libavutil/ppc/types_altivec.h"
#include "libavutil/ppc/util_altivec.h"
#include "libavcodec/hpeldsp.h"
#include "hpeldsp_altivec.h"

#if HAVE_ALTIVEC
/* next one assumes that ((line_size % 16) == 0) */
void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
{
    register vector unsigned char pixelsv1;
    register vector unsigned char pixelsv1B;
    register vector unsigned char pixelsv1C;
    register vector unsigned char pixelsv1D;

    int i;
    register ptrdiff_t line_size_2 = line_size << 1;
    register ptrdiff_t line_size_3 = line_size + line_size_2;
    register ptrdiff_t line_size_4 = line_size << 2;

// hand-unrolling the loop by 4 gains about 15%
// mininum execution time goes from 74 to 60 cycles
// it's faster than -funroll-loops, but using
// -funroll-loops w/ this is bad - 74 cycles again.
// all this is on a 7450, tuning for the 7450
    for (i = 0; i < h; i += 4) {
        pixelsv1  = unaligned_load( 0, pixels);
        pixelsv1B = unaligned_load(line_size, pixels);
        pixelsv1C = unaligned_load(line_size_2, pixels);
        pixelsv1D = unaligned_load(line_size_3, pixels);
        VEC_ST(pixelsv1, 0, (unsigned char*)block);
        VEC_ST(pixelsv1B, line_size, (unsigned char*)block);
        VEC_ST(pixelsv1C, line_size_2, (unsigned char*)block);
        VEC_ST(pixelsv1D, line_size_3, (unsigned char*)block);
        pixels+=line_size_4;
        block +=line_size_4;
    }
}

/* next one assumes that ((line_size % 16) == 0) */
#define op_avg(a,b)  a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) )
void ff_avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
{
    register vector unsigned char pixelsv, blockv;

    int i;
    for (i = 0; i < h; i++) {
        blockv = vec_ld(0, block);
        pixelsv = VEC_LD( 0, pixels);
        blockv = vec_avg(blockv,pixelsv);
        vec_st(blockv, 0, (unsigned char*)block);
        pixels+=line_size;
        block +=line_size;
    }
}

/* next one assumes that ((line_size % 8) == 0) */
static void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
{
    register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
    int i;

   for (i = 0; i < h; i++) {
       /* block is 8 bytes-aligned, so we're either in the
          left block (16 bytes-aligned) or in the right block (not) */
       int rightside = ((unsigned long)block & 0x0000000F);

       blockv = vec_ld(0, block);
       pixelsv = VEC_LD( 0, pixels);

       if (rightside) {
           pixelsv = vec_perm(blockv, pixelsv, vcprm(0,1,s0,s1));
       } else {
           pixelsv = vec_perm(blockv, pixelsv, vcprm(s0,s1,2,3));
       }

       blockv = vec_avg(blockv, pixelsv);

       vec_st(blockv, 0, block);

       pixels += line_size;
       block += line_size;
   }
}

/* next one assumes that ((line_size % 8) == 0) */
static void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
{
    register int i;
    register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
    register vector unsigned char blockv;
    register vector unsigned short pixelssum1, pixelssum2, temp3;
    register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
    register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);

    pixelsv1 = VEC_LD(0, pixels);
    pixelsv2 = VEC_LD(1, pixels);
    pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
    pixelsv2 = VEC_MERGEH(vczero, pixelsv2);

    pixelssum1 = vec_add((vector unsigned short)pixelsv1,
                         (vector unsigned short)pixelsv2);
    pixelssum1 = vec_add(pixelssum1, vctwo);

    for (i = 0; i < h ; i++) {
        int rightside = ((unsigned long)block & 0x0000000F);
        blockv = vec_ld(0, block);

        pixelsv1 = unaligned_load(line_size, pixels);
        pixelsv2 = unaligned_load(line_size+1, pixels);
        pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
        pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
        pixelssum2 = vec_add((vector unsigned short)pixelsv1,
                             (vector unsigned short)pixelsv2);
        temp3 = vec_add(pixelssum1, pixelssum2);
        temp3 = vec_sra(temp3, vctwo);
        pixelssum1 = vec_add(pixelssum2, vctwo);
        pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);

        if (rightside) {
            blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
        } else {
            blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
        }

        vec_st(blockv, 0, block);

        block += line_size;
        pixels += line_size;
    }
}

/* next one assumes that ((line_size % 8) == 0) */
static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
{
    register int i;
    register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
    register vector unsigned char blockv;
    register vector unsigned short pixelssum1, pixelssum2, temp3;
    register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
    register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
    register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);

    pixelsv1 = VEC_LD(0, pixels);
    pixelsv2 = VEC_LD(1, pixels);
    pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
    pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
    pixelssum1 = vec_add((vector unsigned short)pixelsv1,
                         (vector unsigned short)pixelsv2);
    pixelssum1 = vec_add(pixelssum1, vcone);

    for (i = 0; i < h ; i++) {
        int rightside = ((unsigned long)block & 0x0000000F);
        blockv = vec_ld(0, block);

        pixelsv1 = unaligned_load(line_size, pixels);
        pixelsv2 = unaligned_load(line_size+1, pixels);
        pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
        pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
        pixelssum2 = vec_add((vector unsigned short)pixelsv1,
                             (vector unsigned short)pixelsv2);
        temp3 = vec_add(pixelssum1, pixelssum2);
        temp3 = vec_sra(temp3, vctwo);
        pixelssum1 = vec_add(pixelssum2, vcone);
        pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);

        if (rightside) {
            blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
        } else {
            blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
        }

        vec_st(blockv, 0, block);

        block += line_size;
        pixels += line_size;
    }
}

/* next one assumes that ((line_size % 16) == 0) */
static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
{
    register int i;
    register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
    register vector unsigned char blockv;
    register vector unsigned short temp3, temp4,
        pixelssum1, pixelssum2, pixelssum3, pixelssum4;
    register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
    register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);

    pixelsv1 = VEC_LD(0, pixels);
    pixelsv2 = VEC_LD(1, pixels);
    pixelsv3 = VEC_MERGEL(vczero, pixelsv1);
    pixelsv4 = VEC_MERGEL(vczero, pixelsv2);
    pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
    pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
    pixelssum3 = vec_add((vector unsigned short)pixelsv3,
                         (vector unsigned short)pixelsv4);
    pixelssum3 = vec_add(pixelssum3, vctwo);
    pixelssum1 = vec_add((vector unsigned short)pixelsv1,
                         (vector unsigned short)pixelsv2);
    pixelssum1 = vec_add(pixelssum1, vctwo);

    for (i = 0; i < h ; i++) {
        blockv = vec_ld(0, block);

        pixelsv1 = unaligned_load(line_size, pixels);
        pixelsv2 = unaligned_load(line_size+1, pixels);

        pixelsv3 = VEC_MERGEL(vczero, pixelsv1);
        pixelsv4 = VEC_MERGEL(vczero, pixelsv2);
        pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
        pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
        pixelssum4 = vec_add((vector unsigned short)pixelsv3,
                             (vector unsigned short)pixelsv4);
        pixelssum2 = vec_add((vector unsigned short)pixelsv1,
                             (vector unsigned short)pixelsv2);
        temp4 = vec_add(pixelssum3, pixelssum4);
        temp4 = vec_sra(temp4, vctwo);
        temp3 = vec_add(pixelssum1, pixelssum2);
        temp3 = vec_sra(temp3, vctwo);

        pixelssum3 = vec_add(pixelssum4, vctwo);
        pixelssum1 = vec_add(pixelssum2, vctwo);

        blockv = vec_packsu(temp3, temp4);

        vec_st(blockv, 0, block);

        block += line_size;
        pixels += line_size;
    }
}

/* next one assumes that ((line_size % 16) == 0) */
static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
{
    register int i;
    register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
    register vector unsigned char blockv;
    register vector unsigned short temp3, temp4,
        pixelssum1, pixelssum2, pixelssum3, pixelssum4;
    register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
    register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
    register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);

    pixelsv1 = VEC_LD(0, pixels);
    pixelsv2 = VEC_LD(1, pixels);
    pixelsv3 = VEC_MERGEL(vczero, pixelsv1);
    pixelsv4 = VEC_MERGEL(vczero, pixelsv2);
    pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
    pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
    pixelssum3 = vec_add((vector unsigned short)pixelsv3,
                         (vector unsigned short)pixelsv4);
    pixelssum3 = vec_add(pixelssum3, vcone);
    pixelssum1 = vec_add((vector unsigned short)pixelsv1,
                         (vector unsigned short)pixelsv2);
    pixelssum1 = vec_add(pixelssum1, vcone);

    for (i = 0; i < h ; i++) {
        pixelsv1 = unaligned_load(line_size, pixels);
        pixelsv2 = unaligned_load(line_size+1, pixels);

        pixelsv3 = VEC_MERGEL(vczero, pixelsv1);
        pixelsv4 = VEC_MERGEL(vczero, pixelsv2);
        pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
        pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
        pixelssum4 = vec_add((vector unsigned short)pixelsv3,
                             (vector unsigned short)pixelsv4);
        pixelssum2 = vec_add((vector unsigned short)pixelsv1,
                             (vector unsigned short)pixelsv2);
        temp4 = vec_add(pixelssum3, pixelssum4);
        temp4 = vec_sra(temp4, vctwo);
        temp3 = vec_add(pixelssum1, pixelssum2);
        temp3 = vec_sra(temp3, vctwo);

        pixelssum3 = vec_add(pixelssum4, vcone);
        pixelssum1 = vec_add(pixelssum2, vcone);

        blockv = vec_packsu(temp3, temp4);

        VEC_ST(blockv, 0, block);

        block += line_size;
        pixels += line_size;
    }
}

/* next one assumes that ((line_size % 8) == 0) */
static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
{
    register int i;
    register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
    register vector unsigned char blockv, blocktemp;
    register vector unsigned short pixelssum1, pixelssum2, temp3;

    register const vector unsigned char vczero = (const vector unsigned char)
                                        vec_splat_u8(0);
    register const vector unsigned short vctwo = (const vector unsigned short)
                                        vec_splat_u16(2);

    pixelsv1 = VEC_LD(0, pixels);
    pixelsv2 = VEC_LD(1, pixels);
    pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
    pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
    pixelssum1 = vec_add((vector unsigned short)pixelsv1,
                         (vector unsigned short)pixelsv2);
    pixelssum1 = vec_add(pixelssum1, vctwo);

    for (i = 0; i < h ; i++) {
        int rightside = ((unsigned long)block & 0x0000000F);
        blockv = vec_ld(0, block);

        pixelsv1 = unaligned_load(line_size, pixels);
        pixelsv2 = unaligned_load(line_size+1, pixels);

        pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
        pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
        pixelssum2 = vec_add((vector unsigned short)pixelsv1,
                             (vector unsigned short)pixelsv2);
        temp3 = vec_add(pixelssum1, pixelssum2);
        temp3 = vec_sra(temp3, vctwo);
        pixelssum1 = vec_add(pixelssum2, vctwo);
        pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);

        if (rightside) {
            blocktemp = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
        } else {
            blocktemp = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
        }

        blockv = vec_avg(blocktemp, blockv);
        vec_st(blockv, 0, block);

        block += line_size;
        pixels += line_size;
    }
}
#endif /* HAVE_ALTIVEC */

av_cold void ff_hpeldsp_init_ppc(HpelDSPContext *c, int flags)
{
#if HAVE_ALTIVEC
    if (!PPC_ALTIVEC(av_get_cpu_flags()))
        return;

    c->avg_pixels_tab[0][0]        = ff_avg_pixels16_altivec;
    c->avg_pixels_tab[1][0]        = avg_pixels8_altivec;
    c->avg_pixels_tab[1][3]        = avg_pixels8_xy2_altivec;

    c->put_pixels_tab[0][0]        = ff_put_pixels16_altivec;
    c->put_pixels_tab[1][3]        = put_pixels8_xy2_altivec;
    c->put_pixels_tab[0][3]        = put_pixels16_xy2_altivec;

    c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_altivec;
    c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_altivec;
    c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_altivec;
#endif /* HAVE_ALTIVEC */
}
Commit	Line	Data
2ba45a60 DM	1	/*
	2	* Copyright (c) 2002 Brian Foley
	3	* Copyright (c) 2002 Dieter Shirley
	4	* Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org>
	5	*
	6	* This file is part of FFmpeg.
	7	*
	8	* FFmpeg is free software; you can redistribute it and/or
	9	* modify it under the terms of the GNU Lesser General Public
	10	* License as published by the Free Software Foundation; either
	11	* version 2.1 of the License, or (at your option) any later version.
	12	*
	13	* FFmpeg is distributed in the hope that it will be useful,
	14	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	15	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	16	* Lesser General Public License for more details.
	17	*
	18	* You should have received a copy of the GNU Lesser General Public
	19	* License along with FFmpeg; if not, write to the Free Software
	20	* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	21	*/
	22
	23	#include "config.h"
	24
	25	#if HAVE_ALTIVEC_H
	26	#include <altivec.h>
	27	#endif
	28
	29	#include "libavutil/attributes.h"
	30	#include "libavutil/cpu.h"
	31	#include "libavutil/ppc/cpu.h"
	32	#include "libavutil/ppc/types_altivec.h"
	33	#include "libavutil/ppc/util_altivec.h"
	34	#include "libavcodec/hpeldsp.h"
	35	#include "hpeldsp_altivec.h"
	36
	37	#if HAVE_ALTIVEC
	38	/* next one assumes that ((line_size % 16) == 0) */
	39	void ff_put_pixels16_altivec(uint8_t block, const uint8_t pixels, ptrdiff_t line_size, int h)
	40	{
f6fa7814 DM	41	register vector unsigned char pixelsv1;
	42	register vector unsigned char pixelsv1B;
	43	register vector unsigned char pixelsv1C;
	44	register vector unsigned char pixelsv1D;
2ba45a60	45
2ba45a60 DM	46	int i;
	47	register ptrdiff_t line_size_2 = line_size << 1;
	48	register ptrdiff_t line_size_3 = line_size + line_size_2;
	49	register ptrdiff_t line_size_4 = line_size << 2;
	50
	51	// hand-unrolling the loop by 4 gains about 15%
	52	// mininum execution time goes from 74 to 60 cycles
	53	// it's faster than -funroll-loops, but using
	54	// -funroll-loops w/ this is bad - 74 cycles again.
	55	// all this is on a 7450, tuning for the 7450
	56	for (i = 0; i < h; i += 4) {
f6fa7814 DM	57	pixelsv1 = unaligned_load( 0, pixels);
	58	pixelsv1B = unaligned_load(line_size, pixels);
	59	pixelsv1C = unaligned_load(line_size_2, pixels);
	60	pixelsv1D = unaligned_load(line_size_3, pixels);
	61	VEC_ST(pixelsv1, 0, (unsigned char*)block);
	62	VEC_ST(pixelsv1B, line_size, (unsigned char*)block);
	63	VEC_ST(pixelsv1C, line_size_2, (unsigned char*)block);
	64	VEC_ST(pixelsv1D, line_size_3, (unsigned char*)block);
2ba45a60 DM	65	pixels+=line_size_4;
	66	block +=line_size_4;
	67	}
	68	}
	69
	70	/* next one assumes that ((line_size % 16) == 0) */
	71	#define op_avg(a,b) a = ( ((a)\|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) )
	72	void ff_avg_pixels16_altivec(uint8_t block, const uint8_t pixels, ptrdiff_t line_size, int h)
	73	{
f6fa7814	74	register vector unsigned char pixelsv, blockv;
2ba45a60	75
f6fa7814	76	int i;
2ba45a60	77	for (i = 0; i < h; i++) {
2ba45a60	78	blockv = vec_ld(0, block);
f6fa7814	79	pixelsv = VEC_LD( 0, pixels);
2ba45a60 DM	80	blockv = vec_avg(blockv,pixelsv);
	81	vec_st(blockv, 0, (unsigned char*)block);
	82	pixels+=line_size;
	83	block +=line_size;
	84	}
	85	}
	86
	87	/* next one assumes that ((line_size % 8) == 0) */
	88	static void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
	89	{
	90	register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
	91	int i;
	92
	93	for (i = 0; i < h; i++) {
	94	/* block is 8 bytes-aligned, so we're either in the
	95	left block (16 bytes-aligned) or in the right block (not) */
	96	int rightside = ((unsigned long)block & 0x0000000F);
	97
	98	blockv = vec_ld(0, block);
f6fa7814	99	pixelsv = VEC_LD( 0, pixels);
2ba45a60 DM	100
	101	if (rightside) {
	102	pixelsv = vec_perm(blockv, pixelsv, vcprm(0,1,s0,s1));
	103	} else {
	104	pixelsv = vec_perm(blockv, pixelsv, vcprm(s0,s1,2,3));
	105	}
	106
	107	blockv = vec_avg(blockv, pixelsv);
	108
	109	vec_st(blockv, 0, block);
	110
	111	pixels += line_size;
	112	block += line_size;
	113	}
	114	}
	115
	116	/* next one assumes that ((line_size % 8) == 0) */
	117	static void put_pixels8_xy2_altivec(uint8_t block, const uint8_t pixels, ptrdiff_t line_size, int h)
	118	{
	119	register int i;
	120	register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
f6fa7814	121	register vector unsigned char blockv;
2ba45a60 DM	122	register vector unsigned short pixelssum1, pixelssum2, temp3;
	123	register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
	124	register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
	125
f6fa7814 DM	126	pixelsv1 = VEC_LD(0, pixels);
	127	pixelsv2 = VEC_LD(1, pixels);
	128	pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
	129	pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
	130
2ba45a60 DM	131	pixelssum1 = vec_add((vector unsigned short)pixelsv1,
	132	(vector unsigned short)pixelsv2);
	133	pixelssum1 = vec_add(pixelssum1, vctwo);
	134
	135	for (i = 0; i < h ; i++) {
	136	int rightside = ((unsigned long)block & 0x0000000F);
	137	blockv = vec_ld(0, block);
	138
f6fa7814 DM	139	pixelsv1 = unaligned_load(line_size, pixels);
	140	pixelsv2 = unaligned_load(line_size+1, pixels);
	141	pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
	142	pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
2ba45a60 DM	143	pixelssum2 = vec_add((vector unsigned short)pixelsv1,
	144	(vector unsigned short)pixelsv2);
	145	temp3 = vec_add(pixelssum1, pixelssum2);
	146	temp3 = vec_sra(temp3, vctwo);
	147	pixelssum1 = vec_add(pixelssum2, vctwo);
	148	pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
	149
	150	if (rightside) {
	151	blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
	152	} else {
	153	blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
	154	}
	155
	156	vec_st(blockv, 0, block);
	157
	158	block += line_size;
	159	pixels += line_size;
	160	}
	161	}
	162
	163	/* next one assumes that ((line_size % 8) == 0) */
	164	static void put_no_rnd_pixels8_xy2_altivec(uint8_t block, const uint8_t pixels, ptrdiff_t line_size, int h)
	165	{
	166	register int i;
	167	register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
f6fa7814	168	register vector unsigned char blockv;
2ba45a60 DM	169	register vector unsigned short pixelssum1, pixelssum2, temp3;
	170	register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
	171	register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
	172	register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
	173
f6fa7814 DM	174	pixelsv1 = VEC_LD(0, pixels);
	175	pixelsv2 = VEC_LD(1, pixels);
	176	pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
	177	pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
2ba45a60 DM	178	pixelssum1 = vec_add((vector unsigned short)pixelsv1,
	179	(vector unsigned short)pixelsv2);
	180	pixelssum1 = vec_add(pixelssum1, vcone);
	181
	182	for (i = 0; i < h ; i++) {
	183	int rightside = ((unsigned long)block & 0x0000000F);
	184	blockv = vec_ld(0, block);
	185
f6fa7814 DM	186	pixelsv1 = unaligned_load(line_size, pixels);
	187	pixelsv2 = unaligned_load(line_size+1, pixels);
	188	pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
	189	pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
2ba45a60 DM	190	pixelssum2 = vec_add((vector unsigned short)pixelsv1,
	191	(vector unsigned short)pixelsv2);
	192	temp3 = vec_add(pixelssum1, pixelssum2);
	193	temp3 = vec_sra(temp3, vctwo);
	194	pixelssum1 = vec_add(pixelssum2, vcone);
	195	pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
	196
	197	if (rightside) {
	198	blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
	199	} else {
	200	blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
	201	}
	202
	203	vec_st(blockv, 0, block);
	204
	205	block += line_size;
	206	pixels += line_size;
	207	}
	208	}
	209
	210	/* next one assumes that ((line_size % 16) == 0) */
	211	static void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
	212	{
	213	register int i;
	214	register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
f6fa7814	215	register vector unsigned char blockv;
2ba45a60 DM	216	register vector unsigned short temp3, temp4,
	217	pixelssum1, pixelssum2, pixelssum3, pixelssum4;
	218	register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
	219	register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
	220
f6fa7814 DM	221	pixelsv1 = VEC_LD(0, pixels);
	222	pixelsv2 = VEC_LD(1, pixels);
	223	pixelsv3 = VEC_MERGEL(vczero, pixelsv1);
	224	pixelsv4 = VEC_MERGEL(vczero, pixelsv2);
	225	pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
	226	pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
2ba45a60 DM	227	pixelssum3 = vec_add((vector unsigned short)pixelsv3,
	228	(vector unsigned short)pixelsv4);
	229	pixelssum3 = vec_add(pixelssum3, vctwo);
	230	pixelssum1 = vec_add((vector unsigned short)pixelsv1,
	231	(vector unsigned short)pixelsv2);
	232	pixelssum1 = vec_add(pixelssum1, vctwo);
	233
	234	for (i = 0; i < h ; i++) {
	235	blockv = vec_ld(0, block);
	236
f6fa7814 DM	237	pixelsv1 = unaligned_load(line_size, pixels);
f6fa7814 DM	238	pixelsv2 = unaligned_load(line_size+1, pixels);
2ba45a60	239
f6fa7814 DM	240	pixelsv3 = VEC_MERGEL(vczero, pixelsv1);
	241	pixelsv4 = VEC_MERGEL(vczero, pixelsv2);
	242	pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
	243	pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
2ba45a60 DM	244	pixelssum4 = vec_add((vector unsigned short)pixelsv3,
	245	(vector unsigned short)pixelsv4);
	246	pixelssum2 = vec_add((vector unsigned short)pixelsv1,
	247	(vector unsigned short)pixelsv2);
	248	temp4 = vec_add(pixelssum3, pixelssum4);
	249	temp4 = vec_sra(temp4, vctwo);
	250	temp3 = vec_add(pixelssum1, pixelssum2);
	251	temp3 = vec_sra(temp3, vctwo);
	252
	253	pixelssum3 = vec_add(pixelssum4, vctwo);
	254	pixelssum1 = vec_add(pixelssum2, vctwo);
	255
	256	blockv = vec_packsu(temp3, temp4);
	257
	258	vec_st(blockv, 0, block);
	259
	260	block += line_size;
	261	pixels += line_size;
	262	}
	263	}
	264
	265	/* next one assumes that ((line_size % 16) == 0) */
	266	static void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, ptrdiff_t line_size, int h)
	267	{
	268	register int i;
	269	register vector unsigned char pixelsv1, pixelsv2, pixelsv3, pixelsv4;
f6fa7814	270	register vector unsigned char blockv;
2ba45a60 DM	271	register vector unsigned short temp3, temp4,
	272	pixelssum1, pixelssum2, pixelssum3, pixelssum4;
	273	register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
	274	register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
	275	register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
	276
f6fa7814 DM	277	pixelsv1 = VEC_LD(0, pixels);
	278	pixelsv2 = VEC_LD(1, pixels);
	279	pixelsv3 = VEC_MERGEL(vczero, pixelsv1);
	280	pixelsv4 = VEC_MERGEL(vczero, pixelsv2);
	281	pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
	282	pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
2ba45a60 DM	283	pixelssum3 = vec_add((vector unsigned short)pixelsv3,
	284	(vector unsigned short)pixelsv4);
	285	pixelssum3 = vec_add(pixelssum3, vcone);
	286	pixelssum1 = vec_add((vector unsigned short)pixelsv1,
	287	(vector unsigned short)pixelsv2);
	288	pixelssum1 = vec_add(pixelssum1, vcone);
	289
	290	for (i = 0; i < h ; i++) {
f6fa7814 DM	291	pixelsv1 = unaligned_load(line_size, pixels);
f6fa7814 DM	292	pixelsv2 = unaligned_load(line_size+1, pixels);
2ba45a60	293
f6fa7814 DM	294	pixelsv3 = VEC_MERGEL(vczero, pixelsv1);
	295	pixelsv4 = VEC_MERGEL(vczero, pixelsv2);
	296	pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
	297	pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
2ba45a60 DM	298	pixelssum4 = vec_add((vector unsigned short)pixelsv3,
	299	(vector unsigned short)pixelsv4);
	300	pixelssum2 = vec_add((vector unsigned short)pixelsv1,
	301	(vector unsigned short)pixelsv2);
	302	temp4 = vec_add(pixelssum3, pixelssum4);
	303	temp4 = vec_sra(temp4, vctwo);
	304	temp3 = vec_add(pixelssum1, pixelssum2);
	305	temp3 = vec_sra(temp3, vctwo);
	306
	307	pixelssum3 = vec_add(pixelssum4, vcone);
	308	pixelssum1 = vec_add(pixelssum2, vcone);
	309
	310	blockv = vec_packsu(temp3, temp4);
	311
f6fa7814	312	VEC_ST(blockv, 0, block);
2ba45a60 DM	313
	314	block += line_size;
	315	pixels += line_size;
	316	}
	317	}
	318
	319	/* next one assumes that ((line_size % 8) == 0) */
	320	static void avg_pixels8_xy2_altivec(uint8_t block, const uint8_t pixels, ptrdiff_t line_size, int h)
	321	{
	322	register int i;
	323	register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
f6fa7814	324	register vector unsigned char blockv, blocktemp;
2ba45a60 DM	325	register vector unsigned short pixelssum1, pixelssum2, temp3;
	326
	327	register const vector unsigned char vczero = (const vector unsigned char)
	328	vec_splat_u8(0);
	329	register const vector unsigned short vctwo = (const vector unsigned short)
	330	vec_splat_u16(2);
	331
f6fa7814 DM	332	pixelsv1 = VEC_LD(0, pixels);
	333	pixelsv2 = VEC_LD(1, pixels);
	334	pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
	335	pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
2ba45a60 DM	336	pixelssum1 = vec_add((vector unsigned short)pixelsv1,
	337	(vector unsigned short)pixelsv2);
	338	pixelssum1 = vec_add(pixelssum1, vctwo);
	339
	340	for (i = 0; i < h ; i++) {
	341	int rightside = ((unsigned long)block & 0x0000000F);
	342	blockv = vec_ld(0, block);
	343
f6fa7814 DM	344	pixelsv1 = unaligned_load(line_size, pixels);
f6fa7814 DM	345	pixelsv2 = unaligned_load(line_size+1, pixels);
2ba45a60	346
f6fa7814 DM	347	pixelsv1 = VEC_MERGEH(vczero, pixelsv1);
f6fa7814 DM	348	pixelsv2 = VEC_MERGEH(vczero, pixelsv2);
2ba45a60 DM	349	pixelssum2 = vec_add((vector unsigned short)pixelsv1,
	350	(vector unsigned short)pixelsv2);
	351	temp3 = vec_add(pixelssum1, pixelssum2);
	352	temp3 = vec_sra(temp3, vctwo);
	353	pixelssum1 = vec_add(pixelssum2, vctwo);
	354	pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
	355
	356	if (rightside) {
	357	blocktemp = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
	358	} else {
	359	blocktemp = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
	360	}
	361
	362	blockv = vec_avg(blocktemp, blockv);
	363	vec_st(blockv, 0, block);
	364
	365	block += line_size;
	366	pixels += line_size;
	367	}
	368	}
	369	#endif /* HAVE_ALTIVEC */
	370
	371	av_cold void ff_hpeldsp_init_ppc(HpelDSPContext *c, int flags)
	372	{
	373	#if HAVE_ALTIVEC
	374	if (!PPC_ALTIVEC(av_get_cpu_flags()))
	375	return;
	376
	377	c->avg_pixels_tab[0][0] = ff_avg_pixels16_altivec;
	378	c->avg_pixels_tab[1][0] = avg_pixels8_altivec;
	379	c->avg_pixels_tab[1][3] = avg_pixels8_xy2_altivec;
	380
	381	c->put_pixels_tab[0][0] = ff_put_pixels16_altivec;
	382	c->put_pixels_tab[1][3] = put_pixels8_xy2_altivec;
	383	c->put_pixels_tab[0][3] = put_pixels16_xy2_altivec;
	384
	385	c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_altivec;
	386	c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_altivec;
	387	c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_altivec;
	388	#endif /* HAVE_ALTIVEC */
	389	}